Should you believe a high-end graphics card is a must for AI tasks, reconsider. An inventive workaround reveals the potential to transform a $95 AMD CPU into a 16GB GPU, proficient in managing diverse AI workloads on Linux.
The focal processor here is the AMD Ryzen 5 4600G, an affordable APU merging a 6-core, 12-thread Zen 2 CPU with 7-core Radeon Vega integrated graphics. Launched in 2020, it was succeeded by the Ryzen 5 5600G, boasting enhanced CPU prowess yet identical GPU attributes.
Nonetheless, the Ryzen 5 4600G retains unexplored capabilities. A Reddit user, ttio2tech, showcased the APU’s conversion into a 16GB VRAM GPU using a Linux kernel patch and specific software. This patch facilitates system memory access as VRAM, extending beyond the usual 2GB constraint.
This approach enabled the utilization of AI tools like TensorFlow, PyTorch, and JAX on the APU. The AMD Radeon Open Compute platform ROCm software stack, supporting OpenCL and HIP, facilitated this usage. The user affirmed comparable performance to pricier AMD discrete GPUs, such as the Radeon RX 5700 XT and RX 6700 XT, notable for having less VRAM.
The Reddit user, ttio2tech YouTube video showcased the Ryzen 5 4600G successfully managing diverse AI applications such as Stable Diffusion, MiniGPT-4, FastChat, Alpaca-LoRA, Whisper, LLaMA, and LLM. Regrettably, detailed insights on configuring the Ryzen 5 4600G with AI software on Linux remain absent. The YouTuber has committed to releasing a comprehensive setup guide.
Regarding performance, the Ryzen 5 4600G impressively generated a 512 x 512-pixel image in just about one minute and 50 seconds, using default settings of 50 steps. This achievement rivals some high-end processors, proving exceptional for a $95 APU. Although DDR4 memory was employed, specifications were omitted. Notably, while the Ryzen 5 4600G natively supports DDR4-3200, many instances can reach DDR4-4000, sparking curiosity about AI performance scaling with faster memory.
The user aimed to democratize AI, granting access to those without means for expensive GPUs. The goal also involved urging AMD to officially endorse and simplify this feature’s activation. This innovative hack exemplifies repurposing older hardware through creativity. It underscores the substantial AI potential within AMD APUs, given their notable memory bandwidth and efficient power consumption.
CEO
Aleem Iqbal serves as CEO of Red Awakening. He holds an ICS from Degree College and a BS in Computer Science from Virtual University. An avid gamer for over 15 years and technology enthusiast, Aleem brings a unique perspective shaped by his diverse background. His passion for gaming led him to become an author and tech leader. Aleem leverages his experience to create engaging stories and lead Red Awakening towards future growth.
